swarm_sdk 2.7.7 → 2.7.9

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: b83fd113813e0667b02d12c5ef6b9f4d47c134afa0665e2b6e482a6c9d3dacd7
-  data.tar.gz: 8a758e06817af7690c21fc446597b0e1a91b876643d20a4a111a62bc953d3d5e
+  metadata.gz: d6af3c1d2b4f71135747902e3d94b4541904bf1877c02064cb928cccc742cf9b
+  data.tar.gz: a829be669e85d25c1fa85cc05410b44bb3d9785ad0a72bfe238ae92f1859c828
 SHA512:
-  metadata.gz: 5bfcfaf82ca105e6b1e1f04e009c703b641c25af6797ccd72c945a0c8dbbfbe60a0d783fc9e8d43322bbefac95b16123c4e7232c4ac9e1850a770d1120568344
-  data.tar.gz: 0aa16f8d6ebe0fd2acb4cb52baba570000c82c4607d015d3a4569d671f9743cb1199edd80953002b8dc0fb673fb464ebfc2d5d0c645f3afb50258c7451281304
+  metadata.gz: af53d4fa1e8a4ae2ace5297e90743ddd219e3015f3ffa62e89dedb607f73e9641d3cb3d44b506a76e2a70a72074aa3f504229cb4a7d02b056af6b0b555c59bdd
+  data.tar.gz: 333359fe2c10fbb772675ae035fa00de4eb1a1cf9d3f5ee7e7b4440d8d6bd25cfc96d1efbadf8621e5bf3c6ef53f8e790a8d31db3c682c1ac7f47d61507c4843

data/lib/swarm_sdk/agent/llm_instrumentation_middleware.rb

@@ -52,23 +52,23 @@ module SwarmSDK
         # Execute request
         @app.call(env).on_complete do |response_env|
           end_time = Time.now
-          duration = end_time - start_time
+
+          # Determine if this was a streaming request based on whether chunks were accumulated
+          # This is more reliable than parsing response content
+          is_streaming = accumulated_raw_chunks.any?
 
           # For streaming: use accumulated raw SSE chunks
           # For non-streaming: use response body
-          raw_body = if accumulated_raw_chunks.any?
-            accumulated_raw_chunks.join
-          else
-            response_env.body
-          end
+          raw_body = is_streaming ? accumulated_raw_chunks.join : response_env.body
 
           # Store SSE body in Fiber-local for citation extraction
           # This allows append_citations_to_content to access the full SSE body
           # even though response.body is empty for streaming responses
-          Fiber[:last_sse_body] = raw_body if accumulated_raw_chunks.any?
+          Fiber[:last_sse_body] = raw_body if is_streaming
 
           # Emit response event
-          emit_response_event(response_env, start_time, end_time, duration, raw_body)
+          timing = { start_time: start_time, end_time: end_time, duration: end_time - start_time }
+          emit_response_event(response_env, timing, raw_body, is_streaming)
         end
       end
 
@@ -96,21 +96,17 @@ module SwarmSDK
       # Emit response event
       #
       # @param env [Faraday::Env] Response environment
-      # @param start_time [Time] Request start time
-      # @param end_time [Time] Request end time
-      # @param duration [Float] Request duration in seconds
+      # @param timing [Hash] Timing information with :start_time, :end_time, :duration keys
       # @param raw_body [String, nil] Raw response body (SSE stream for streaming, JSON for non-streaming)
+      # @param streaming [Boolean] Whether this was a streaming response (determined by chunk accumulation)
       # @return [void]
-      def emit_response_event(env, start_time, end_time, duration, raw_body)
-        # Detect if this is a streaming response (starts with "data:")
-        streaming = raw_body.is_a?(String) && raw_body.start_with?("data:")
-
+      def emit_response_event(env, timing, raw_body, streaming)
         response_data = {
           provider: @provider_name,
           body: parse_body(raw_body),
           streaming: streaming,
-          duration_seconds: duration.round(3),
-          timestamp: end_time.utc.iso8601,
+          duration_seconds: timing[:duration].round(3),
+          timestamp: timing[:end_time].utc.iso8601,
           status: env.status,
         }
 
@@ -166,6 +162,9 @@ module SwarmSDK
 
       # Parse request/response body
       #
+      # For requests: returns parsed JSON hash
+      # For responses: returns full body (JSON parsed or raw string for SSE)
+      #
       # @param body [String, Hash, nil] HTTP body
       # @return [Hash, String, nil] Parsed body
       def parse_body(body)
@@ -177,8 +176,9 @@ module SwarmSDK
         # Try to parse JSON
         JSON.parse(body)
       rescue JSON::ParserError
-        # Return truncated string if not JSON
-        body.to_s[0..1000]
+        # Return full body for SSE/non-JSON responses
+        # Don't truncate - let consumers decide how to handle large bodies
+        body.to_s
       rescue StandardError
         nil
       end

data/lib/swarm_sdk/config.rb

@@ -92,6 +92,7 @@ module SwarmSDK
       mcp_log_level: ["SWARM_SDK_MCP_LOG_LEVEL", -> { Defaults::Logging::MCP_LOG_LEVEL }],
       default_execution_timeout: ["SWARM_SDK_DEFAULT_EXECUTION_TIMEOUT", -> { Defaults::Timeouts::EXECUTION_TIMEOUT_SECONDS }],
       default_turn_timeout: ["SWARM_SDK_DEFAULT_TURN_TIMEOUT", -> { Defaults::Timeouts::TURN_TIMEOUT_SECONDS }],
+      mcp_request_timeout: ["SWARM_SDK_MCP_REQUEST_TIMEOUT", -> { Defaults::Timeouts::MCP_REQUEST_SECONDS }],
     }.freeze
 
     # WebFetch and control settings

data/lib/swarm_sdk/defaults.rb

@@ -94,6 +94,47 @@ module SwarmSDK
       # Time-to-live for cached response IDs. 5 minutes allows conversation
       # continuity while preventing stale cache issues.
       RESPONSES_API_TTL_SECONDS = 300
+
+      # MCP client request timeout (seconds)
+      #
+      # Default timeout for MCP server connections. 5 minutes accommodates
+      # long-running SSE streams and tool executions. This timeout applies to
+      # the entire operation (operation_timeout in HTTPX), so it must be long
+      # enough for SSE connections that may run for extended periods.
+      MCP_REQUEST_SECONDS = 300
+    end
+
+    # MCP reconnection configuration
+    #
+    # Settings for automatic reconnection when SSE/streamable connections drop.
+    # Note: The background SSE notification stream uses operation_timeout which
+    # limits total connection duration. Since this stream is meant to stay open
+    # indefinitely for server notifications, we configure aggressive reconnection
+    # so timeouts are transparent to users. Tool calls use separate connections
+    # and are unaffected by SSE stream timeouts.
+    module McpReconnection
+      # Maximum number of reconnection attempts
+      #
+      # Very high value (effectively infinite) because the SSE notification stream
+      # is expected to timeout periodically due to operation_timeout limitations.
+      # Reconnection is transparent - tool calls continue working regardless.
+      MAX_RETRIES = 1000
+
+      # Initial delay between reconnection attempts (milliseconds)
+      #
+      # Fast initial reconnect (500ms) to minimize notification gaps.
+      INITIAL_DELAY_MS = 500
+
+      # Exponential backoff growth factor
+      #
+      # Slow growth (1.2x) because we expect frequent reconnections.
+      # 500ms -> 600ms -> 720ms -> 864ms -> 1037ms -> ...
+      DELAY_GROW_FACTOR = 1.2
+
+      # Maximum delay between reconnection attempts (milliseconds)
+      #
+      # Caps at 10 seconds to ensure responsive reconnection even after many retries.
+      MAX_DELAY_MS = 10_000
     end
 
     # Output and content size limits